parallel feap 8.2

[tolis1981]

 I tried to install the parallel version in one gentos linux cluster machine.I followed the  parallel installation manual and I made install all the libraries.I made install the parfeap and I received this message 
 make: [install] error 1 (ignored)
The parfeap executable was not created.
Does anyone has an idea what went wrong?

[FEAP_Admin]

(1) Do you have a working version of PETSc that you have independently tested with Fortran?

assuming that

(2) You will need to determine which file is throwing the compile error.

With Feap 8.2 you the source is very far behind relative to PETSc revisions.  If you search the user forum you will find
pointers on updating the source to be compatible with current releases of PETSc.  Note, that even the makefile needs updating.  Unfortunately there are have been so many updates to PETSc since Feap 8.2 was released there is no way for me to just give you a list of needed changes.

[tolis1981]

The PETSc version that I use is v.2.3.1, which is  compatible  to the 8.2 version.
I downloaded and I installed it and I run the (make test all).I can not understand which is the compiler error.
unix/pmetis.o: In function `pmetis_':
pmetis.f:(.text+0x73): undefined reference to `metis_partgraphkway_'
pmetis.f:(.text+0xf2): undefined reference to `metis_partgraphrecursive_'
usolve.o: In function `usolve_':
/home/****/Ver82/parfeap/usolve.F:232: undefined reference to `pcsetco                                                                                                                                                             ordinates_'
collect2: ld returned 1 exit status
make: [install] Error 1 (ignored)

[FEAP_Admin]

It looks like you do not have the metis libraries installed together with your petsc libraries.   Can you check this.

[tolis1981]

Thank you for your response.
There are the parMetis (parmetis-4.0.3.tar.gz) and metis (metis-5.1.0.tar.gz),   (mt-metis-0.4.4.tar.gz) which one is better to use?

[FEAP_Admin]

For parFEAP you only need metis.
However if you are building metis via petsc I recommend parmetis and metis.  We typicially build this into our petsc configuration.  If you do it separately, then you will need to manually add the -l flags to the final link line.

Here is our current configuration line for petsc:

./configure --download-{parmetis,superlu_dist,mpich,ml,hypre,metis,mumps,scalapack,blacs} --with-debugging=0

Note we no longer have Prometheus in the configure as it has been superseded by GAMG in newer versions of petsc.

[tolis1981]

Thank you for your answer.
I tried to use the command you told me in Petsc 2.3.0- 2.3.1 (which is compatible with parfeap8.2) and it had many problems in installing the newer versions of the metis, parmetis etc with download.I tried to do it manually but for some reason It cannot see the directory path that I give to it as valid.
I download petsc 3.3.5 and the command worked ok but the feap 8.2 becomes crazy when I try to compile it.It does not built a lot of objects files.
Now I have a dilemma:Which one is better to fix the petsc or the source files of feap?
Which files I have to change in order to make it work in any case?

[FEAP_Admin]

You are a lot of Feap versions behind so it is hard to say how many changes you will have to make to bring your code up-to-date in terms of petsc release -- there are even (minor) changes to be made in the makefile.  If you scan through the discussion forum's petsc board you will find most of the corrections mentioned.  But essentially you will have to look at each compile error and fix it.  It is tedious but not too difficult.

[tolis1981]

Thank you for your quick response.
So you advice me to change the feap code and makefile rather than to download old versions of the parmetis, metis and the other libraries and manually install them.
Just in case do you know how to install them maually?What am I doing wrong?

[FEAP_Admin]

getting old version to work can be difficult; they often will not compile with new compilers among other problems.  updating your FEAP will be tedious but likely easier.